What's new in Android machine learning
summary
Android adds concrete on-device ML advances: TensorFlow Lite in Google Play services (BETA) reduces app binary size by ~2 MB and teams see up to 20% CPU wins, MLKit reaches >1 billion MAU and 80k apps, and Google Play-served TF Lite runs 400M MAU doing ~20 billion daily inferences. Hardware acceleration yields up to 43% lower pose latency on GPU, TPU cuts object/pose latency 28–50%, DSP makes inference 3x faster at 1/7 the energy. Text Recognition V2 covers 37 languages (~4.9B people) with ~30% faster recognition. Acceleration Service uses local benchmarking to pick stable, fast accelerators; updatable NN API and Play Services delegates roll out.
AI-written summary · about
chapters (8)
- 0:00Intro
- 0:51Magic Eraser
- 3:07GPU Pose Detection (Beta)
- 5:20ML Kit Early Access Program
- 6:23Essentials built-in
- 8:49Service for Android
- 10:34Updatable Hardware Drivers
- 11:04TensorFlow Lite in Google Play Services
related talks
43mWhat’s new in Android machine learning
10mWhat's new in Android Machine Learning
47mMyths and Reality of Android Machine Learning
39mMachine Learning on Android Demystified
43mMyths and Reality of Android Machine Learning
10mMachine Learning for mobile
21mHow to use hardware acceleration for Machine Learning inference on Android
47mMachine Learning on Android Demystified